Closed AllanElleuch closed 1 year ago
Hey @AllanElleuch, thanks for the detailed report.
Yes, the way you have it there with the volume setup (-v ~/root-cache:/root/.cache
), it will only download the model once (and you'll be able to see them in ~/root-cache
on the host system). Not sure what's up with HuggingFace, maybe they were / are having some temporary issues.
It's also possible to bundle the model with the container (as in https://github.com/kiri-art/docker-diffusers-api-build-download), but that's usually only useful for serverless, since you're able to persist the volume on your local machine.
But either way, you still have to download the model once (but only once).
If HuggingFace is playing up, you could try add the following callInput
:
MODEL_URL="https://pub-bdad4fdd97ac4830945e90ed16298864.r2.dev/diffusers/models--runwayml--stable-diffusion-v1-5--fp16.tar.zst"
and it should download from kiri's public R2 storage (rate-limited).
Let me know how that goes.
Thanks Gadicc for the detailed feedback.
It seems that the issue was an unstable connection on my side. Now it is working perfectly fine and I was able to download all the files without an issue!
Fantastic, thanks for reporting back, @AllanElleuch. Happy Diffusing 🧨 :tada:
Hi,
I started locally an instance and run a generation request from the documentation. However I receive a timeout that seems to be caused by the fact that the request timeout from the hugging face CDN.
Pulling 15 files take times, is it possible to warmup the server and download all this before? Or keep the download instead of dropping them.
Thanks!
Docker command
Request
Server log
Postman side